Piani, Mirko
(2026)
Geospatial assessment of fruit growth and quality through the implementation of a computer vision system, [Dissertation thesis], Alma Mater Studiorum Università di Bologna.
Dottorato di ricerca in
Scienze e tecnologie agrarie, ambientali e alimentari, 38 Ciclo.
Documenti full-text disponibili:
Abstract
Advances in precision orchard management increasingly rely on high‑resolution, scalable, and automated methods for monitoring fruit growth and quality. Traditional approaches—based on manual measurements or specialised sensing devices—remain limited by labour intensity, cost, and restricted sampling capacity. This thesis develops and evaluates a suite of computer vision–based methodologies designed to enable geospatial, non‑destructive, and high‑throughput assessment of fruit size and color using consumer‑grade imaging systems. The first part of the work introduces a 2D fruit sizing algorithm that integrates distance normalization with local image entropy to improve boundary detection under variable lighting, phenological stages, and camera–fruit distances. The method achieves a mean absolute error of approximately 3 mm and demonstrates robustness across viewpoints and developmental stages. Building on this foundation, the second part optimises the sizing pipeline for near‑real‑time applications by analyzing fruit morphological allometries and identifying intrinsic geometric uncertainty thresholds. A redesigned, CPU‑efficient implementation reduces per‑fruit processing time to 1 ms while maintaining high accuracy, enabling the analyzis of up to 63 fruit per frame. The third component compares monocular and stereo RGB‑D systems for orchard scale fruit quality assessment. Results show that stereo imagery provides more reliable depth‑based sizing and crop load estimation, whereas monocular systems—enhanced by deep learning–based depth inference—offer high‑frequency monitoring and superior color rendering for daily quality tracking. Finally, the thesis presents a trunk‑based geospatial data collection framework that integrates autonomous vehicles, IoT sensors, and manual measurements into a unified time‑series database, enabling plant‑level mapping and long‑term monitoring across heterogeneous orchard environments. Together, these contributions demonstrate how lightweight computer vision algorithms, depth‑aware imaging, and geospatial data integration can support scalable, cost‑effective, and biologically informed monitoring of fruit growth and quality.
Abstract
Advances in precision orchard management increasingly rely on high‑resolution, scalable, and automated methods for monitoring fruit growth and quality. Traditional approaches—based on manual measurements or specialised sensing devices—remain limited by labour intensity, cost, and restricted sampling capacity. This thesis develops and evaluates a suite of computer vision–based methodologies designed to enable geospatial, non‑destructive, and high‑throughput assessment of fruit size and color using consumer‑grade imaging systems. The first part of the work introduces a 2D fruit sizing algorithm that integrates distance normalization with local image entropy to improve boundary detection under variable lighting, phenological stages, and camera–fruit distances. The method achieves a mean absolute error of approximately 3 mm and demonstrates robustness across viewpoints and developmental stages. Building on this foundation, the second part optimises the sizing pipeline for near‑real‑time applications by analyzing fruit morphological allometries and identifying intrinsic geometric uncertainty thresholds. A redesigned, CPU‑efficient implementation reduces per‑fruit processing time to 1 ms while maintaining high accuracy, enabling the analyzis of up to 63 fruit per frame. The third component compares monocular and stereo RGB‑D systems for orchard scale fruit quality assessment. Results show that stereo imagery provides more reliable depth‑based sizing and crop load estimation, whereas monocular systems—enhanced by deep learning–based depth inference—offer high‑frequency monitoring and superior color rendering for daily quality tracking. Finally, the thesis presents a trunk‑based geospatial data collection framework that integrates autonomous vehicles, IoT sensors, and manual measurements into a unified time‑series database, enabling plant‑level mapping and long‑term monitoring across heterogeneous orchard environments. Together, these contributions demonstrate how lightweight computer vision algorithms, depth‑aware imaging, and geospatial data integration can support scalable, cost‑effective, and biologically informed monitoring of fruit growth and quality.
Tipologia del documento
Tesi di dottorato
Autore
Piani, Mirko
Supervisore
Co-supervisore
Dottorato di ricerca
Ciclo
38
Coordinatore
Settore disciplinare
Settore concorsuale
Parole chiave
Computer vision, fruit sizing, precision orchard management, color estimation, near-real-time, trunk-based-localization, precision horticulture
Data di discussione
23 Luglio 2026
URI
Altri metadati
Tipologia del documento
Tesi di dottorato
Autore
Piani, Mirko
Supervisore
Co-supervisore
Dottorato di ricerca
Ciclo
38
Coordinatore
Settore disciplinare
Settore concorsuale
Parole chiave
Computer vision, fruit sizing, precision orchard management, color estimation, near-real-time, trunk-based-localization, precision horticulture
Data di discussione
23 Luglio 2026
URI
Gestione del documento: